Forecasting Cloud Spend: A Case Study

Forecasting Cloud Cost

The Background

With many businesses moving from on-premise IT infrastructure to cloud-based infrastructure, spending on cloud services has increased significantly. The cloud migration has only intensified over the last year and many IT organizations wish to have better predictability with regards to cloud spend.

Apart from a pure budgeting standpoint, better visibility into the future cloud spend can help organizations to put forward effective checks and balances around the usage of cloud resources.

Moreover, the eventual spending heavily depends on the split between committed usage and pay-as-you-go usage. Whereas committed usage is heavily discounted pay-as-you-go models can turn out to be quite expensive. Organizations with a better grasp of future usage can simply save millions of dollars by optimally configuring these parameters or by negotiating better enterprise agreements.

Our Customer 

Our customer is one of the largest cloud users for one of the top three cloud service providers, with an annual cloud spend of well over 25 million dollars.


Lack of internal protocols and guidelines around usage had made the cloud cost spiral up. At the same time, it had been difficult to set such guidelines without proper visibility into the future

Committing on usage had been difficult for the customer, hence they had committed on lower than optimal levels to avoid wastage

The Approach

Spends on each service were modeled and forecasted.

The forecasting frequency was monthly and the forecasts were generated 12 months into the future.

Each forecasting model was a multivariate model that incorporated the above given factors as predictor variables.

The total spend forecast was obtained by simply adding up the forecast for each of the services.

18 months of historical data was used for forecasting 

Forecasting cloud spend using ML models

A typical forecasting process considers historical trends and seasonality. However, by incorporating factors that influence the cloud expenditure into the forecast, we can obtain higher accuracies. Following are the factors considered.

  • Users: The number of users in development and testing environments
  • Usage: Usage in terms of time & volume in resources
  • Projects: Number of projects that consumed cloud resources
  • Holidays & special offers: Events that can increase the number of transactions in the production environment
  • Commitments: Predefined committed usage for computing resources

Forecasting Accuracies: backtesting results

93%   January 2021

89%   February 2021

90%   March 2021


Using FORECAST², the customer was able to forecast spending at a high level of accuracy.

The forecasting accuracy for each of the service forecasted varied between 70% – 98%

The overall accuracy (sum of all services) was between 89% – 93% at the monthly level

When aggregated to three months, the forecast was 95% accurate

The Outcome 

The customer has started negotiating an enterprise-wide long-term contract based on forecasts provided by FORECAST² 

  • The customer has come up with a set of guidelines with appropriate checks and balances to manage cloud cost without hampering productivity.
  • The customer is in the process of negotiating a long-term contract with the provider with optimal commitment levels. This exercise is expected to reduce the cloud cost by 2 -5 million USD.
  • The customer has started budgeting the cloud cost with a more accurate and reliable view of the future.

To know more about how your business can benefit from using Forecasting at Scale, please reach out to us or simply sign up for a free trial.

Sri Lanka
No. 63/1,
Dhrmapala Mawatha,
Sri Lanka.
No. 9, Raffles Place,
Republic Plaza,
Singapore, 048619.
Menara Digitaraya,
Jl. Raden Saleh Raya,
No. 46A, Jakarta,